BLOCK SCHEDULING OF ITERATIVE ALGORITHMS AND GRAPH-LEVEL PRIORITY SCHEDULING IN A SIMULATED DATA-FLOW MULTIPROCESSOR

While data-flow principles permit the utilization of large-scale multiprocessor systems with high programmability and good efficiency, they also introduce much overhead at runtime. In this paper, we have studied an important class of PDE solvers, namely iterative methods for solving linear systems. Although these methods are inherently highly sequential, we have found that much parallelism could be exploited in a data-flow system by scheduling the iterative part of the algorithms in blocks and by looking ahead across several iterations. This approach is general and will apply to other iterative and loop-based problems. It is also demonstrated by simulation means, that relying solely on data-driven scheduling of parallel and unrolled loops results in low resource utilization and poor performance. A graph-level priority scheduling mechanism has been developed that drastically improves resource utilization and yields higher performance.